Robust Optimal Design of an Autopilot for Transitional Flights of a Tail-sitter Mini Uav

The transitional flight of tail-sitter aircraft is a challenging problem because the flight includes a wide operating range with nonlinearity and at low speed approaches the stall. In order to solve this problem, a robust autopilot design method using a variable environment genetic algorithm (VE-GA) is proposed. VE-GA is a new robust optimization method based on a real coded genetic algorithm (RCGA). Here, the word “environment” refers to the uncertainties considered in the evaluation functions. In a VEGA, the environment is changed repeatedly after several generations. In this manner, genes go through many types of environments over generations, and obtain robustness against uncertainties. In order to improve the efficiency and accuracy of the optimization, we introduce a local optimization method—Powell's direction set method (PDSM), and term the combined robust optimization procedure VE-GA/PDSM. Finally, our proposed method is applied to the offline-based parameter optimization of a neural network (NN) which is part of a tailsitter mini unmanned aerial vehicle’s (UAV) autopilot architecture.